Policymakers should think of self-driving cars as an integral part of the transportation equation, not separate from it.

Share story

THE human obsession with self-driving cars goes back a long way. General Motors unveiled the concept of autonomous cars in the 1939 World’s Fair. Since then, self-driving cars have become fixtures in sci-fi movies, emblematic of technological and social progress.

This fiction, very rapidly, is becoming fact. Even three to four years ago, when Google started experimenting with self-driving cars, it seemed that it would be a while before that technology hit the road. Things couldn’t have been further from reality.

I recently rode to work in Uber’s self-driving car. I even passed Tesla’s almost-autonomous Model S. However, the leap from “almost” to “fully” autonomous car needs to overcome an abyss of technical and regulatory challenges.

The National Highway Traffic Safety Administration has defined five levels of vehicle automation. Level 0 implies that the driver is in complete control of the vehicle and Level 5 means complete autonomy — the driver is not expected to operate the vehicle at any time.

Most cars on the road today have Level 1 automation. Tesla, with its “autopilot,” is the only production car that has Level 3 automation. Level 5 automation has been difficult to achieve due to several factors, many of them technical.

The most important issue is that self-driving cars haven’t been driven enough. Google’s cars have logged only 1.5 million miles of autonomous driving, which is just two to three times of what an average American drives in a lifetime. Tesla has been taking a faster approach by accumulating data through all its autopilot customers and has racked up 780,000 miles, 100,000 of them self-driven.

Self-driving, at its core, is a machine-learning problem. The more varied data these companies can collect, the sooner they can get to Level 4 autonomy, which means the driver can still take back control of the car in certain conditions.

It is not surprising Uber chose Pittsburgh. The city, with its crazy lane changes, ever-changing weather and potholed roads, is a test bed for its driverless cars due to its “double black diamond of driving” distinction, not to mention the world-class robotics talent at Carnegie Mellon University. The university has been involved in autonomous vehicles for quite some time: In 1995, CMU research scientist Dean Pomerleau and student Todd Jochem took the “No hands across America trip.” The 2,849-mile cross-country trip in a Pontiac minivan was 98.7 percent self-driven, a culmination of a decade of CMU research. The university also won the DARPA urban challenge in 2007, which involved the Tartan car driving autonomously through 55 miles of tough city conditions.

Currently, self-driving cars are unable to deal with all weather conditions. Sensors, using sonar, cameras and LIDAR, have difficulty working in extreme weather. Companies are installing new sensors to add redundancy and make cars fail-proof. However, adding more sensors comes at a cost. The Google AV module is currently priced at $80,000.

Self-driving cars are also unable to deal with potholes and construction. Typically this problem arises because maps are not updated in real time, which can throw the car into a quandary. The rapid focus on autonomous cars has fueled investments in better mapping technology. Ford recently invested in Civil Maps to generate high-resolution 3D maps, and BMW, Daimler and Audi jointly acquired Nokia’s digital-mapping company HERE for $3.1 billion.

What is the bar that self-driving cars need to meet to achieve market and regulatory approval?”

Another challenge is human drivers. In the foreseeable future, the road network will be shared between human- and self-driven cars. Will the self-driven cars understand the nuances of human behavior? More important, will human driving become less accident prone in the presence of self-driven cars? These questions still remain unanswered.

A big resistance to self-driving cars might come from the labor market. Close to 1.5 million truck drivers and several hundred thousand cabdrivers would lose their jobs in the U.S. when self-driving is a reality. This will definitely have a huge impact on policy and costs associated with self-driving cars.

In addition to these issues, one big question remains: safety. What is the bar that self-driving cars need to meet to achieve market and regulatory approval?

There has been only one verified case of Tesla’s autopilot leading to a crash. And although the Google car has been involved in a few accidents, it was at fault only in one minor crash. Although more data are required to conclusively prove that self-driving cars are better than human drivers, the evidence is mounting. Especially when distracted driving is on the rise.

Today, the fatality of driving stands at 1.08 per 100 million miles. Is that a reasonable benchmark for self-driving cars? Or do we expect them to do much better? How much time will it take us to get there?

Tesla claims it will have Level 5 autonomous cars on the roads in the next two years, whereas Google hints it may be 2020. In any case, it is safe to say that autonomous cars will not completely take over in the next five to 10 years. Although, automobile manufacturers might not be up for it, sharing data across different platforms might hasten the maturity of this technology and prevent road accidents during testing.

One question that is staring transit planners in the face is: How to design public transportation for the next 20 to 30 years in the presence of self-driving technology? In preliminary research at Carnegie Mellon University’s Heinz College, we found that although Uber can perform as a partial substitute for public transportation in dense metropolitan areas, it acts more as a complement in sparse urban areas and suburbs. People can ride transit to work and some may opt to ride home using Uber. Policymakers should think of self-driving cars as an integral part of the transportation equation, not separate from it.

Mass transit will remain, ecologically and economically, the better alternative. However, self-driving cars can solve the “last-mile” problem and address the needs of people not served well by public transit.

On Sept. 20, the National Highway Safety Transportation Administration released a 15-point guideline for autonomous cars that aims to create a public dialogue on adoption of self-driving cars. As technology and policy converge, the question is not if self-driving cars will come to market, but when.